Methodology
How the portfolios are built, tested and costed — and how the Research Lab tests signals and picks individual ideas. Plain English, no proprietary detail.
The portfolios
Eight portfolios, spanning a deliberate spectrum from a plain cheap baseline to an aggressive momentum basket. All are long-only and shown out-of-sample from January 2019; most rebalance monthly, while the Free-Cash-Flow Value screen rebalances annually:
- Simple Core. The honest baseline — a cheap global index + government bonds + money-market blend at fixed weights. No optimisation. The reference everything else is measured against; the point is that a cheap index-and-bonds mix is hard to beat.
- Adaptive Conservative. A tactical cross-asset portfolio that each month selects the most attractive risk-adjusted assets from a global menu (equity, bonds, gold, commodities, managed-futures trend), sized to ~7% volatility and holding a large cash buffer when little qualifies — the highest risk-adjusted return with the smallest drawdowns. It concentrates rather than holding everything at once, so it is adaptive, not a static "all-weather" mix.
- Adaptive Balanced. The same tactical engine at ~10% volatility — concentrates in what is working and de-risks into cash in stress, beating a 60/40 on risk-adjusted return at roughly half the drawdown of equities.
- Global Momentum. Each month holds the ten best-performing global equity names by trailing 12-month return, inverse-vol weighted. Higher return and a larger drawdown — the "ride the winners" showcase, diversified across ten names.
- Growth. Momentum, toned down — roughly 60% in the momentum-winners sleeve and the rest in a money-market buffer, for much of the momentum return with a gentler drawdown.
- FCF Value. Our free-cash-flow study put to work — the ten S&P 500 names cheapest on free cash flow relative to enterprise value, rebalanced annually.
- US (S&P 500 anchor) and UK (FTSE 100 anchor). Take a single index as the anchor and add uncorrelated diversifiers — "improve the index you already hold".
The universe
The investment universe is around 905 instruments. It splits into two tiers based on which broker the strategy is being run for:
- Available on both HL and IB Pro (~780 instruments): UK and EU-listed UCITS ETFs (Vanguard, iShares, Invesco, Xtrackers, WisdomTree, SPDR); commodity ETCs (gold, silver, oil, agricultural baskets, individual softs/metals); single stocks from the S&P 500, FTSE 100 / 250, and EuroStoxx 50; US-listed single-stock ADRs of major Asian and Latin American companies; UK investment trusts (multi-asset, infrastructure, private equity, renewables, hedge-fund-style trusts like BH Macro and Pershing Square); bond ETFs across UK, US, and EU government and corporate curves.
- IB Pro only (~100 instruments): US-listed ETFs blocked on HL retail under PRIIPs/KID rules — managed-futures trend products (DBMF, KMLM, WTMF), US REITs (VNQ, IYR, SCHH), the SPDR Select sector suite (XLU, XLP, XLV, XLE, XLF, XLI, XLY, XLK, XLB), factor ETFs (USMV, MTUM, QUAL, VLUE, SIZE, SPLV), currency trusts (UUP, FXE, FXY, FXB, FXF), tail-hedge / vol products (TAIL, VIXY, PUTW, BTAL), the deeper US Treasury curve (IEF, SHY, TLH, BIL, AGG), gold (GLD, IAU), broad commodities (DBC, GSG), inflation-linked bonds (TIP), and international equity (VEA, VWO, EFA).
This split is structural — it's the same maths either way, but the IB Pro run gets to consider a wider set of genuinely uncorrelated diversifiers (notably US-listed gold, managed futures, and currency trusts) that HL retail simply cannot hold.
Each product uses its own estimation lookback — from ~2 years for the diversified cross-asset products up to ~6 years for the index-anchored ones. A name must have enough clean history within that window to be ranked, which keeps freshly-listed funds from being judged on a few months of noisy data. Instruments with corrupted price data (any single-day move >50%, almost always a corporate-action artefact) are excluded automatically, and only currently-tradeable instruments can ever be held (no delisted or closed funds appear as picks).
Walk-forward, out-of-sample
The performance shown is out-of-sample by construction. At every rebalance point the portfolio is built using only data that was available before that date. The next week's returns are then earned on those weights with no further intervention. This is the strictest out-of-sample design — it simulates exactly what an investor would have experienced in real time, with no hindsight.
The track record runs from January 2019 to July 2026 (7.5 years) — a single common out-of-sample start date used across every product so the curves are directly comparable on the same window, including the 2020 and 2022 stress periods.
Rebalancing
The default cadence is monthly, on the first trading day of each calendar month — typically the 1st (or the next business day if the 1st falls on a weekend or bank holiday). Trades are assumed to execute at that day's closing price, giving an investor a full trading session to place the orders during normal UK market hours.
A weekly cadence is also available on each product page — rebalances then fall on the first trading day of each calendar week (typically Monday). Weekly trades the portfolio ~52 times a year vs ~12 for monthly; on a £20,000 portfolio that extra dealing turnover materially increases costs and is rarely worth the responsiveness it buys. Between rebalances the weights drift naturally with market moves and are not touched.
A small drift filter avoids tiny trades: if a position has not drifted more than 5 percentage points from its new target, it is held rather than nudged back into line. This applies to every type of holding (ETFs, ETCs and shares) and meaningfully reduces dealing fees and FX cost.
Costs — Retail ISA (HL) vs Professional (IB Pro)
Every cost an investor would actually pay is deducted from the equity curve. We run the strategy under two account profiles side-by-side: a Retail Stocks & Shares ISA at Hargreaves Lansdown (tax-free wrapper, restricted universe, expensive trades) and a Professional account at Interactive Brokers UK Pro (full universe, institutional-grade execution).
| Cost | Hargreaves Lansdown | Interactive Brokers UK Pro |
|---|---|---|
| Dealing fee (ETF / share) | £11.95 / trade (£5.95 frequent-trader) | ~£0.80 / trade ($1 USD min) |
| FX cost on non-GBP trades | ~1.00% (HL retail tier) | ~0.002% (~2 bps spread) |
| Bid-offer spread (ETFs) | ~5 bps per side | ~2 bps per side (tighter) |
| Platform fee (S&S ISA) | 0.45% annually, ETF cap £45/yr | £0 — no platform fee |
| Underlying fund OCF | Varies by fund — pass-through | Varies by fund — pass-through |
| Universe accessible | UCITS ETFs + LSE singles + International Dealing for foreign singles. US-listed ETFs blocked for UK retail (PRIIPs) | Full universe — including US-listed ETFs (managed futures, US REITs, sector pure-plays, currency trusts, vol products) |
At the default £20,000 initial investment, total fee drag on Hargreaves runs ~50–250 bps/yr depending on how much non-GBP trading the strategy ends up doing; on IB Pro it's typically ~10–40 bps/yr. The gap narrows for larger portfolios (the per-trade fees become a smaller share of NAV) and widens for smaller ones. The dual equity curves on each product page show the actual delta for that strategy.
Why we model both
The product is the same strategy in either case — the same diversification thesis, the same rebalance cadence, the same picks. But the cost structures are so different that the realised investor return is materially different. We think it's misleading to publish a single equity curve without telling you which platform that curve assumes. So we publish both, and let you pick the one that fits your situation.
What is not modelled
The simulation is fair but the following real-world frictions are not included. A live investor would experience them and should expect realised returns to fall short of the simulation by some amount as a result.
- Execution slippage. Trades are assumed to fill at the day's closing price. A real order may move the market against you, especially for smaller AUM positions in illiquid trusts.
- Cash-flow timing. No model of contributions, withdrawals, or cash-drag from settlement delays (T+2, FX settlement).
- Tax events outside the ISA wrapper. Inside an ISA, gains and dividends are tax-free. Outside an ISA, CGT and dividend tax would apply and reduce realised performance.
- Broker outages or order rejections. Each rebalance assumes the trade actually executed on the intended day; in practice this could slip.
- Survivorship bias in the universe. The candidate set is built largely from instruments that exist today, so companies that delisted, merged or left an index over the period are under-represented. We mitigate this where it matters — the Momentum product selects winners (the names that delist are typically losers it would not have held), and in-data delisting is rare in our set — but a true point-in-time index membership is not reconstructed. Treat results as the survivor-set upper bound.
- Dividend reinvestment and withholding tax. Total-return proxies are used where available; some single stocks are price-only and so understate true total return.
Research Lab — testing signals and picking ideas
The Research Lab is the second arm: an evidence-first framework for deciding which investment signals actually predict returns, and which individual ideas earn a place. The principles:
- Evidence over narrative. A good story isn't a reason to buy. A claim earns its place only with evidence that it predicts returns — tested on our own data, in recent windows, net of cost — and we keep the ones that fail on the record too.
- Test honestly. The right benchmark (a small-cap signal measured against small caps, not the S&P), no look-ahead (enter when the information is public, not when it was created), honest survivorship, and conservative (week-clustered) standard errors. We design the test so it can prove us wrong.
- Grade decisively. Every signal gets a graded scorecard — effect size, significance, robustness across regimes and benchmarks, durability, fit to our universe — and a one-line verdict, not a vague "it depends."
- Signals are overlays, not bets. Most real signals are short-lived — good for timing an entry, weak as the sole reason to hold for a year. Durable strands (quality, value, capital allocation) carry a position; signals sit on top.
- Single ideas: margin of safety first. We act only when the price is clearly below a defensible value; size by conviction with the reasoning recorded; and never re-size, re-price or re-date a position after the fact — the record shows what we actually did, not a flattering rewrite.
- Publish the misses. Negative results and rejected ideas are kept and tracked, so the record isn't flattered by deletion.
The graded signal studies and the plain-English concepts behind them live in the Research Lab; live ideas are tracked in the £20,000 book there. Research and education — not advice.
Important reminders
This material is an academic research project and is published for general information and educational purposes only. It does not constitute investment advice, a financial promotion, or a recommendation to buy, sell, or hold any security. The hypothetical performance shown has not been achieved by any real investor. Past performance is not a reliable indicator of future returns. Capital is at risk. The value of investments can fall as well as rise. Tax treatment depends on individual circumstances and may change.
This site is not authorised or regulated by the Financial Conduct Authority. If you are considering investing, please consult an independent financial adviser authorised and regulated by the FCA.